Lemma EC.3, p. ec7 — tight distributions F̄_L, F̄_M, F̄_H attaining (EC.5)–(EC.7) (corrected: (a) for D > 0; (c) last piece closed at S)
ProvedValuePersPricing.Deviation.lemma_EC_3lambert-wp2o-batch-pfp2ap2o-correctedp2o-gran-per-chapterp2o-plan-paperp2o-v1personalized-pricingtight-bound
Let , take cost , and let be the thresholds at . Each statement below asserts the existence of a probability law of a valuation with mean , scale and coefficient of deviation whose cCDF is, for , the stated function, and which attains the corresponding bound with equality.
- Low heterogeneity. For , with ,
and (Eq. (EC.5) with equality). 2. Medium heterogeneity. For , with ,
and (Eq. (EC.6) with equality). 3. High heterogeneity. For , with ,
and (Eq. (EC.7) with equality).
These laws show that the bounds of Theorem 2 are tight in the standardized case; Lemma 1 transports them to a general cost and mean.
Differences from the source:
- correction: (a) "Suppose D ∈ [0, δL]" → D ∈ (0, δL]; reason: at D = 0, α_L = 1 and F̄_L is the point mass at 1, whose scale is 1, not S > 1 (e.g. S = 2).
- correction: (c) third piece "α_H/x if x ∈ (α_H/(1−D), S)", "0 otherwise" → α_H/x on (α_H/(1−D), S]; reason: as printed F̄_H(S) = 0 while F̄_H(x) → α_H/S > 0 as x ↑ S, but x ↦ P(V ≥ x) is left-continuous, so no law has the printed cCDF; the mean computation (EC.13) and the optimal prices [α_H/(1−D), S] need the atom at S.
- encoding: the cCDF on its whole domain → compared on x ≥ 0 only; reason: "0 otherwise" would also cover x < 0, where P(V ≥ x) = 1 for V ≥ 0.
- encoding: "Eq. (EC.6) with equality" → D·R_PP = log(S)·R_SP; reason: equivalent since D > 0 on [δL, δM].
- encoding: "a random variable V" → a probability measure ν on ℝ with Market ν 0; S > 1 as a hypothesis; reason: standing assumptions of §2 and of Theorem 2.
Preamble
import Mathlib import Definitions.Def_ValuePersPricing_Deviation_Model open MeasureTheory Set Filter Topology
Formal statement
namespace ValuePersPricing.Deviation
/-- Lemma EC.3 (p. ec7): tight distributions for `c = 0`, `µ = 1`, `S > 1` (δ's at `M = 1`).
Each part gives a law with the stated cCDF on `x ≥ 0`, scale `S`, coefficient of deviation `D`
and mean `1` that attains (EC.5), (EC.6), (EC.7) respectively with equality.
Corrected: (a) for `D ∈ (0, δ_L]` (at `D = 0` the printed law is the point mass at 1, of scale 1);
(c) the third piece of `F̄_H` is on `(α_H/(1 − D), S]` (the printed `F̄_H(S) = 0` is not
left-continuous, so no law has it). -/
theorem lemma_EC_3 (S : ℝ) (hS : 1 < S) :
(∀ D ∈ Ioc 0 (deltaL 1 S),
∃ ν : Measure ℝ, Market ν 0 ∧ mean ν = 1 ∧ IsScale ν S ∧ coefDev ν = D ∧
(∀ x : ℝ, 0 ≤ x → tail ν x =
if x < (lambertWm1 ((D - 1) / Real.exp 1) / (D - 1))⁻¹ then 1
else if x ≤ 1 then (lambertWm1 ((D - 1) / Real.exp 1) / (D - 1))⁻¹ / x
else if x ≤ S then D / (Real.log S * x)
else 0) ∧
ppProfit ν 0 =
lambertWm1 ((D - 1) / Real.exp 1) / (D - 1) * spProfit ν 0) ∧
(∀ D ∈ Icc (deltaL 1 S) (deltaM 1 S),
∃ ν : Measure ℝ, Market ν 0 ∧ mean ν = 1 ∧ IsScale ν S ∧ coefDev ν = D ∧
(∀ x : ℝ, 0 ≤ x → tail ν x =
if x = 0 then 1
else if x < Real.exp 1 * S ^ (1 - 1 / D) then
D / Real.log S / Real.exp 1 * S ^ (1 / D - 1)
else if x ≤ S then D / Real.log S / x
else 0) ∧
coefDev ν * ppProfit ν 0 = Real.log S * spProfit ν 0) ∧
(∀ D ∈ Icc (deltaM 1 S) (deltaH 1 S),
∃ ν : Measure ℝ, Market ν 0 ∧ mean ν = 1 ∧ IsScale ν S ∧ coefDev ν = D ∧
(∀ x : ℝ, 0 ≤ x → tail ν x =
if x = 0 then 1
else if x ≤ (-lambertWm1 (-1 / (Real.exp 1 * S * (1 - D))))⁻¹ / (1 - D) then 1 - D
else if x ≤ S then (-lambertWm1 (-1 / (Real.exp 1 * S * (1 - D))))⁻¹ / x
else 0) ∧
ppProfit ν 0 =
-lambertWm1 (-1 / (Real.exp 1 * S * (1 - D))) * spProfit ν 0) := by sorry
end ValuePersPricing.Deviation
Source
Elmachtoub, Gupta & Hamilton, The Value of Personalized Pricing, SSRN 3127719 (review version, PDF sha256 635e22eb…), p. ec7 (PDF 39), Lemma EC.3; proof pp. ec8–ec9 (PDF 40–41)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.